| { |
| "@context": { |
| "@language": "en", |
| "@vocab": "https://schema.org/", |
| "sc": "https://schema.org/", |
| "cr": "http://mlcommons.org/croissant/", |
| "rai": "http://mlcommons.org/croissant/RAI/", |
| "dct": "http://purl.org/dc/terms/", |
| "prov": "http://www.w3.org/ns/prov#", |
| "conformsTo": "dct:conformsTo", |
| "citeAs": "cr:citeAs", |
| "recordSet": "cr:recordSet", |
| "field": "cr:field", |
| "dataType": { |
| "@id": "cr:dataType", |
| "@type": "@vocab" |
| }, |
| "source": "cr:source", |
| "fileObject": "cr:fileObject", |
| "fileSet": "cr:fileSet", |
| "extract": "cr:extract", |
| "jsonPath": "cr:jsonPath", |
| "containedIn": "cr:containedIn", |
| "includes": "cr:includes" |
| }, |
| "@type": "sc:Dataset", |
| "name": "SysCON3D", |
| "alternateName": [ |
| "syscon3d", |
| "syscon3d-neurips26/syscon3d" |
| ], |
| "description": "SysCON3D is a deterministic benchmark bundle for stress-testing multi-view 3D reconstruction backbones and 3D consistency metrics. It contains Mip-NeRF 360 reference images, calibration split manifests, and materialized inconsistent image sets including cross-scene mixtures, one-outlier samples, identical-image samples, Gaussian noise, patched Gaussian corruptions, and small Gaussian perturbations of otherwise consistent views.", |
| "url": "https://huggingface.co/datasets/syscon3d-neurips26/syscon3d", |
| "license": "https://huggingface.co/datasets/syscon3d-neurips26/syscon3d#license-and-source-data", |
| "conformsTo": [ |
| "http://mlcommons.org/croissant/1.1", |
| "http://mlcommons.org/croissant/RAI/1.0" |
| ], |
| "version": "6", |
| "citeAs": "SysCON3D anonymous NeurIPS submission, 2026.", |
| "datePublished": "2026-05-07", |
| "creator": { |
| "@type": "sc:Organization", |
| "name": "Anonymous authors" |
| }, |
| "keywords": [ |
| "3d reconstruction", |
| "multi-view consistency", |
| "benchmark", |
| "Mip-NeRF 360", |
| "robustness", |
| "Croissant" |
| ], |
| "distribution": [ |
| { |
| "@type": "cr:FileObject", |
| "@id": "readme", |
| "name": "README.md", |
| "description": "Dataset card with usage, extraction, source-data, and license notes.", |
| "contentUrl": "https://huggingface.co/datasets/syscon3d-neurips26/syscon3d/resolve/main/README.md", |
| "encodingFormat": "text/markdown", |
| "contentSize": "2568", |
| "sha256": "5249a3dc6025c3efb402e85fe0a0ae78bb9bca3224dfccfb483042df1cd6598f" |
| }, |
| { |
| "@type": "cr:FileObject", |
| "@id": "calibration_manifest", |
| "name": "mipnerf360_calibration_splits.json", |
| "description": "Consistent-scene calibration split manifest.", |
| "contentUrl": "https://huggingface.co/datasets/syscon3d-neurips26/syscon3d/resolve/main/mipnerf360_calibration_splits.json", |
| "encodingFormat": "application/json", |
| "contentSize": "11071", |
| "sha256": "a9696fbce1f3c31ecd450db06687727c95a97ff4a7f0241c50498c830f541536" |
| }, |
| { |
| "@type": "cr:FileObject", |
| "@id": "impossible_manifest", |
| "name": "mipnerf360_impossible_splits.json", |
| "description": "SysCON3D stress-test split manifest with deterministic sample ids and image paths.", |
| "contentUrl": "https://huggingface.co/datasets/syscon3d-neurips26/syscon3d/resolve/main/mipnerf360_impossible_splits.json", |
| "encodingFormat": "application/json", |
| "contentSize": "934871", |
| "sha256": "926d7b6eecbbcc5116d5422050802bd964e5cf92b93e136c6cd60914e72997e6" |
| }, |
| { |
| "@type": "cr:FileObject", |
| "@id": "mipnerf360_archive_000", |
| "name": "archives/syscon3d_mipnerf360_000.tar", |
| "description": "Uncompressed tar shard containing the portable mipnerf360/ payload referenced by the manifests.", |
| "contentUrl": "https://huggingface.co/datasets/syscon3d-neurips26/syscon3d/resolve/main/archives/syscon3d_mipnerf360_000.tar", |
| "encodingFormat": "application/x-tar", |
| "contentSize": "659394560", |
| "sha256": "78f20b3efc4b394162e20e3206a08122aa3a1d67f86a98632f7fcd3662afb06d" |
| }, |
| { |
| "@type": "cr:FileSet", |
| "@id": "materialized_syscon3d_images", |
| "name": "Materialized SysCON3D stress-test images", |
| "description": "Deterministic 224x224 PNG images for the materialized inconsistent scene types.", |
| "containedIn": { |
| "@id": "mipnerf360_archive_000" |
| }, |
| "includes": "mipnerf360/syscon3d_scene_types/**/*.png", |
| "encodingFormat": "image/png" |
| }, |
| { |
| "@type": "cr:FileSet", |
| "@id": "referenced_mipnerf360_images", |
| "name": "Referenced Mip-NeRF 360 images", |
| "description": "Referenced source images needed by the calibration splits and portable manifests.", |
| "containedIn": { |
| "@id": "mipnerf360_archive_000" |
| }, |
| "includes": "mipnerf360/*/images_4/*", |
| "encodingFormat": "image/jpeg" |
| }, |
| { |
| "@type": "cr:FileSet", |
| "@id": "camera_metadata", |
| "name": "Camera metadata", |
| "description": "Per-scene transform metadata for the referenced Mip-NeRF 360 scenes.", |
| "containedIn": { |
| "@id": "mipnerf360_archive_000" |
| }, |
| "includes": "mipnerf360/*/transforms.json", |
| "encodingFormat": "application/json" |
| } |
| ], |
| "recordSet": [ |
| { |
| "@type": "cr:RecordSet", |
| "@id": "syscon3d_stress_test_samples", |
| "name": "SysCON3D stress-test samples", |
| "description": "Samples listed by scene type in mipnerf360_impossible_splits.json.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "stress_test/sample_id", |
| "name": "sample_id", |
| "description": "Deterministic sample identifier.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { |
| "@id": "impossible_manifest" |
| }, |
| "extract": { |
| "jsonPath": "$.*[*].sample_id" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "stress_test/subset_size", |
| "name": "subset_size", |
| "description": "Number of views in the multi-view sample.", |
| "dataType": "sc:Integer", |
| "source": { |
| "fileObject": { |
| "@id": "impossible_manifest" |
| }, |
| "extract": { |
| "jsonPath": "$.*[*].subset_size" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "stress_test/image_rel_paths", |
| "name": "image_rel_paths", |
| "description": "Image paths relative to the extracted mipnerf360/ dataset root.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { |
| "@id": "impossible_manifest" |
| }, |
| "extract": { |
| "jsonPath": "$.*[*].image_rel_paths" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "stress_test/source_scenes", |
| "name": "source_scenes", |
| "description": "Underlying source scene names used to construct each sample.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { |
| "@id": "impossible_manifest" |
| }, |
| "extract": { |
| "jsonPath": "$.*[*].source_scenes" |
| } |
| } |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "@id": "syscon3d_calibration_splits", |
| "name": "SysCON3D calibration splits", |
| "description": "Consistent-scene calibration splits listed in mipnerf360_calibration_splits.json.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "calibration/scenes", |
| "name": "scenes", |
| "description": "Source scene names included in the calibration manifest.", |
| "dataType": "sc:Text", |
| "source": { |
| "fileObject": { |
| "@id": "calibration_manifest" |
| }, |
| "extract": { |
| "jsonPath": "$.scenes" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "calibration/subset_sizes", |
| "name": "subset_sizes", |
| "description": "View counts used by the calibration manifest.", |
| "dataType": "sc:Integer", |
| "source": { |
| "fileObject": { |
| "@id": "calibration_manifest" |
| }, |
| "extract": { |
| "jsonPath": "$.subset_sizes" |
| } |
| } |
| } |
| ] |
| } |
| ], |
| "rai:dataLimitations": [ |
| "SysCON3D is designed for stress-testing multi-view 3D reconstruction backbones and 3D consistency metrics. It is not intended as a general-purpose training dataset, semantic recognition benchmark, or substitute for real deployment evaluation.", |
| "Coverage is limited to nine static Mip-NeRF 360 scenes, deterministic image corruptions, fixed view counts, and 224x224 materialized stress-test images. Results may not generalize to dynamic scenes, human-centered scenes, outdoor-only or indoor-only deployment domains, or non-photographic imagery." |
| ], |
| "rai:dataBiases": [ |
| "The source scenes inherit the selection biases of Mip-NeRF 360, including a small number of mostly static real-world scenes and specific camera trajectories.", |
| "The inconsistent samples intentionally over-represent synthetic and adversarial stress cases such as cross-scene mixtures and Gaussian corruptions; these samples are not representative of naturally occurring multi-view captures." |
| ], |
| "rai:personalSensitiveInformation": "The benchmark is based on public scene photographs and does not intentionally collect personal or sensitive attributes. It may still contain incidental real-world background content inherited from the source images.", |
| "rai:dataUseCases": [ |
| "Recommended: evaluating robustness and abstention behavior of multi-view 3D reconstruction backbones and 3D consistency metrics under controlled stress tests.", |
| "Not recommended: training production models, evaluating demographic fairness, evaluating semantic recognition, or making claims about safety outside the documented stress-test setting." |
| ], |
| "rai:dataSocialImpact": "The benchmark can improve transparency around failure modes of learned 3D reconstruction backbones and metrics. Misuse risk includes overclaiming robustness beyond the documented scenes and perturbations or treating synthetic stress-test behavior as equivalent to real-world safety.", |
| "rai:hasSyntheticData": true, |
| "rai:dataCollection": "Source photographs and camera metadata come from the Mip-NeRF 360 benchmark. SysCON3D selects referenced images and materializes deterministic stress-test samples from those sources plus synthetic image corruptions.", |
| "rai:dataPreprocessingProtocol": "The release uses referenced-only packaging, rewrites manifests to portable paths under mipnerf360/, and stores materialized stress-test PNGs at 224x224. Synthetic scene types are generated deterministically from recorded sample ids, seeds, source paths, and corruption parameters in mipnerf360_impossible_splits.json.", |
| "rai:dataAnnotationProtocol": "No human semantic labels are included. The manifests provide programmatic sample metadata such as sample id, scene type, view count, source scenes, source image paths, synthetic seeds, and corruption parameters.", |
| "rai:dataReleaseMaintenancePlan": "The anonymous review release is versioned by the manifest field version=6 and by the Hugging Face dataset commit. Future updates should increment the manifest version and preserve prior release artifacts when possible.", |
| "prov:wasDerivedFrom": [ |
| { |
| "@id": "https://jonbarron.info/mipnerf360/", |
| "name": "Mip-NeRF 360" |
| } |
| ], |
| "prov:wasGeneratedBy": [ |
| { |
| "@type": "prov:Activity", |
| "name": "SysCON3D materialization", |
| "description": "Deterministic construction of calibration splits, cross-scene mixtures, identical-image samples, one-outlier samples, Gaussian noise samples, patched Gaussian samples, and Gaussian perturbations of consistent image sets." |
| } |
| ] |
| } |
|
|
